Analytics

The Analytics page aggregates request logs into trends, answering "how has everything been running lately".

What to look at

Dimension	Use
Request volume	Which models are being used, and which client is using them.
Success rate	Whether the whole thing is stable.
Latency distribution	Whether responses are slowing down, and where.
Token usage	Which logical model and channel consume the most.
Failover frequency	Which provider keeps dropping the ball.
Failure-cause distribution	The share of timeout / rate limit / 5xx / auth.

Division of labor with request logs

Request logs are per-request, used to pinpoint what happened to one request.
Analytics is aggregate, used to spot patterns over a period.

It analyzes your own requests

The statistics here come only from local request logs, and are a different thing from anonymous usage telemetry. The latter is covered in Local-first & secrets.

Troubleshooting approach

Read together with failover, typical calls:

A provider's failure rate spikes → look at its cooldown records and channel diagnostics.
Success rate is fine but latency rises → a slow candidate may be pulling the average up; consider adjusting queue order.
Token usage is abnormal → check Cache affinity to confirm whether the prompt cache still hits.

Runtime logs record what the service itself is doing; see Runtime logs.